sdk/emulator/qemu.git
8 years agosun4uv: fix serial initialization regression
Marc-André Lureau [Sat, 22 Oct 2016 09:52:44 +0000 (12:52 +0300)]
sun4uv: fix serial initialization regression

Since commit b6607a1a204d, serial_hds_isa_init() was introduced to
factor out serial_isa_init() loops. However, sun4uv shouldn't start from
0 when there is a mm serial on 0 already. Add a "from" argument to
serial_hds_isa_init().

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20161022095318.17775-5-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoringbuf: fix chr_write return value
Marc-André Lureau [Sat, 22 Oct 2016 09:52:43 +0000 (12:52 +0300)]
ringbuf: fix chr_write return value

It should return the number of written bytes.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20161022095318.17775-4-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agochar: remove use-after-free on win-stdio
Marc-André Lureau [Sat, 22 Oct 2016 09:52:42 +0000 (12:52 +0300)]
char: remove use-after-free on win-stdio

Found by reviewing the code, win_stdio_close() is called by
qemu_chr_free() which then call qemu_chr_free_common() taking care of
freeing CharDriverState*.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20161022095318.17775-3-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agorng: remove unused included header
Marc-André Lureau [Sat, 22 Oct 2016 09:52:41 +0000 (12:52 +0300)]
rng: remove unused included header

DEFINE_PROP_CHR is not used (rng is not of TYPE_DEVICE)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20161022095318.17775-2-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agochar.h: misc doc fix
Marc-André Lureau [Tue, 11 Oct 2016 15:20:12 +0000 (19:20 +0400)]
char.h: misc doc fix

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20161011152012.3228-1-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agochar: serial: check divider value against baud base
Prasad J Pandit [Wed, 12 Oct 2016 05:58:08 +0000 (11:28 +0530)]
char: serial: check divider value against baud base

16550A UART device uses an oscillator to generate frequencies
(baud base), which decide communication speed. This speed could
be changed by dividing it by a divider. If the divider is
greater than the baud base, speed is set to zero, leading to a
divide by zero error. Add check to avoid it.

Reported-by: Huawei PSIRT <psirt@huawei.com>
Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
Message-Id: <1476251888-20238-1-git-send-email-ppandit@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agomemory: optimize memory_region_sync_dirty_bitmap
Paolo Bonzini [Fri, 23 Sep 2016 09:08:54 +0000 (11:08 +0200)]
memory: optimize memory_region_sync_dirty_bitmap

Avoid walking the FlatView of all address spaces.  Most of the
address spaces will have no log_sync callback on their listeners.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agomemory: optimize memory_global_dirty_log_sync
Paolo Bonzini [Thu, 22 Sep 2016 14:09:08 +0000 (16:09 +0200)]
memory: optimize memory_global_dirty_log_sync

Only return a nonzero dirty_log_mask for RAM/ROM memory regions.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agomemory: add a per-AddressSpace list of listeners
Paolo Bonzini [Thu, 22 Sep 2016 14:23:06 +0000 (16:23 +0200)]
memory: add a per-AddressSpace list of listeners

This speeds up MEMORY_LISTENER_CALL noticeably.  Right now,
with many PCI devices you have N regions added to M AddressSpaces
(M = # PCI devices with bus-master enabled) and each call looks
up the whole listener list, with at least M listeners in it.
Because most of the regions in N are BARs, which are also roughly
proportional to M, the whole thing is O(M^3).  This changes it
to O(M^2), which is the best we can do without rewriting the
whole thing.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agomemory: eliminate global MemoryListeners
Paolo Bonzini [Thu, 22 Sep 2016 14:11:54 +0000 (16:11 +0200)]
memory: eliminate global MemoryListeners

There is none, so just drop the code.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agotcg: try sti when moving a constant into a dead memory temp
Paolo Bonzini [Thu, 15 Sep 2016 13:16:00 +0000 (15:16 +0200)]
tcg: try sti when moving a constant into a dead memory temp

This comes from free from unifying tcg_reg_alloc_mov and
tcg_reg_alloc_movi's handling of TEMP_VAL_CONST.  It triggers
often on moves to cc_dst, such as the following translation
of "sub $0x3c,%esp":

  before:                          after:
  subl   $0x3c,%ebp                subl   $0x3c,%ebp
  movl   %ebp,0x10(%r14)           movl   %ebp,0x10(%r14)
  movl   $0x3c,%ebx                movl   $0x3c,0x2c(%r14)
  movl   %ebx,0x2c(%r14)

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1473945360-13663-1-git-send-email-pbonzini@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agotarget-i386: fix 32-bit addresses in LEA
Paolo Bonzini [Wed, 12 Oct 2016 07:23:39 +0000 (09:23 +0200)]
target-i386: fix 32-bit addresses in LEA

This was found with test-i386.  The issue is that instructions
such as

    addr32 lea (%eax), %rax

did not perform a 32-bit extension, because the LEA translation
skipped the gen_lea_v_seg step.  That step does not just add
segments, it also takes care of extending from address size to
pointer size.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agotest-i386: fix bitrot for 64-bit
Paolo Bonzini [Wed, 12 Oct 2016 06:57:19 +0000 (08:57 +0200)]
test-i386: fix bitrot for 64-bit

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoqht-bench: relax test_start/stop atomic accesses
Emilio G. Cota [Fri, 14 Oct 2016 09:54:51 +0000 (11:54 +0200)]
qht-bench: relax test_start/stop atomic accesses

test_start/stop are used only as flags to loop on. Barriers are unnecessary,
since no dependent data is transferred among threads apart from the flags
themselves.

This commit relaxes the three accesses to test_start/stop that were
not yet relaxed.

Signed-off-by: Emilio G. Cota <cota@braap.org>
8 years agoatomic: base mb_read/mb_set on load-acquire and store-release
Paolo Bonzini [Mon, 19 Sep 2016 09:36:44 +0000 (11:36 +0200)]
atomic: base mb_read/mb_set on load-acquire and store-release

This introduces load-acquire and store-release operations in QEMU.
For now, just use them as an implementation detail of atomic_mb_read
and atomic_mb_set.

Since docs/atomics.txt documents that atomic_mb_read only synchronizes
with an atomic_mb_set of the same variable, we can use the new implementation
everywhere instead of seq-cst loads and stores.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agorcu: simplify memory barriers
Paolo Bonzini [Mon, 19 Sep 2016 09:27:46 +0000 (11:27 +0200)]
rcu: simplify memory barriers

Thanks to the acquire semantics of qemu_event_reset and qemu_event_wait,
some memory barriers can be removed.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoqemu-thread: use acquire/release to clarify semantics of QemuEvent
Paolo Bonzini [Mon, 19 Sep 2016 09:10:57 +0000 (11:10 +0200)]
qemu-thread: use acquire/release to clarify semantics of QemuEvent

Do not use the somewhat mysterious atomic_mb_read/atomic_mb_set,
instead make sure that the operations on QemuEvent are annotated
with the desired acquire and release semantics.

In particular, qemu_event_set wakes up the waiting thread, so it must
be a release from the POV of the waker (compare with qemu_mutex_unlock).
And it actually needs a full barrier, because that's the only thing that
provides something like a "load-release".

Use smp_mb_acquire until we have atomic_load_acquire and
atomic_store_release in atomic.h.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoatomic: introduce smp_mb_acquire and smp_mb_release
Paolo Bonzini [Mon, 19 Sep 2016 08:50:38 +0000 (10:50 +0200)]
atomic: introduce smp_mb_acquire and smp_mb_release

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoPut the copyright information on a separate line
Thomas Huth [Wed, 5 Oct 2016 09:54:44 +0000 (11:54 +0200)]
Put the copyright information on a separate line

The output string QEMU with "--version" is very long, it does
not fit into a normal line of a terminal window anymore. By
putting the copyright information on a separate line instead,
the output looks much nicer.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1475661284-30153-1-git-send-email-thuth@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoblock/iscsi: Adding new iSER transport layer option
Roy Shterman [Sun, 9 Oct 2016 08:14:56 +0000 (11:14 +0300)]
block/iscsi: Adding new iSER transport layer option

iSER is a new transport layer supported in Libiscsi,
iSER provides a zero-copy RDMA capable interface that can
improve performance.

In order to use the new iSER transport one need to have RDMA supported HW
and to choose iser as the protocol name in Libiscsi URI.

For now iSER memory buffers are pre-allocated and pre-registered,
hence in order to work with iSER from QEMU, one need to enable
MEMLOCK attribute in the VM to be large enough for all iSER buffers and RDMA
resources.

Signed-off-by: Roy Shterman <roysh@mellanox.com>
Message-Id: <1476000896-18632-3-git-send-email-roysh@mellanox.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoblock/iscsi: Introducing new zero-copy API
Roy Shterman [Sun, 9 Oct 2016 08:14:55 +0000 (11:14 +0300)]
block/iscsi: Introducing new zero-copy API

A new API to deploy zero-copy command submission. The new API takes I/O
vectors list and number of I/O vectors to submit as input parameters
when initiating the command. New API must be used if working with
iSER transport option.

Signed-off-by: Roy Shterman <roysh@mellanox.com>
Message-Id: <1476000896-18632-2-git-send-email-roysh@mellanox.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agorbd: shift byte count as a 64-bit value
Paolo Bonzini [Mon, 10 Oct 2016 19:58:58 +0000 (21:58 +0200)]
rbd: shift byte count as a 64-bit value

Otherwise, reads of more than 2GB fail.  Until commit
7bbca9e290a9c7c217b5a24fc6094e91e54bd05d, reads of 2^41
bytes succeeded at least theoretically.

In fact, pdiscard ought to receive a 64-bit integer as the
count for the same reason.

Reported by Coverity.

Fixes: 7bbca9e290a9c7c217b5a24fc6094e91e54bd05d
Cc: qemu-stable@nongnu.org
Cc: kwolf@redhat.com
Cc: eblake@redhat.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agokvm-all: don't use stale dbg_data->cpu
Alex Bennée [Mon, 10 Oct 2016 15:46:25 +0000 (16:46 +0100)]
kvm-all: don't use stale dbg_data->cpu

The changes to run_on_cpu and friends mean that all helpers are passed
the CPUState of vCPU they are running on. The conversion missed the
field in commit e0eeb4a21a3ca4b296220ce4449d8acef9de9049 which
introduced bugs.

Reported-by: Claudio Imbrenda <imbrenda@linux.vnet.ibm.com>
Tested-by: Claudio Imbrenda <imbrenda@linux.vnet.ibm.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20161010154625.14881-1-alex.bennee@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/riku/tags/pull-linux-user-20160921' into staging
Peter Maydell [Fri, 21 Oct 2016 12:49:58 +0000 (13:49 +0100)]
Merge remote-tracking branch 'remotes/riku/tags/pull-linux-user-20160921' into staging

Linux-user changes, mostly bugfixes and adding support for some
new syscalls and some obscure syscalls as well. Includes some
missed patches from earlier rounds, and dropping unicore32 target.

v2: fix the syslog patch and test build with clang-3.8
v3: drop ustat patch

# gpg: Signature made Fri 21 Oct 2016 13:38:06 BST
# gpg:                using RSA key 0xB44890DEDE3C9BC0
# gpg: Good signature from "Riku Voipio <riku.voipio@iki.fi>"
# gpg:                 aka "Riku Voipio <riku.voipio@linaro.org>"
# Primary key fingerprint: FF82 03C8 C391 98AE 0581  41EF B448 90DE DE3C 9BC0

* remotes/riku/tags/pull-linux-user-20160921: (21 commits)
  linux-user: disable unicore32 linux-user build
  linux-user: added support for pwritev() system call.
  linux-user: added support for preadv() system call.
  linux-user: Fix fadvise64() syscall support for Mips32
  linux-user: Redirect termbits.h for Mips64 to termbits.h for Mips32
  linux-user: Update ioctls definitions for Mips32
  linux-user: Update mips_syscall_args[] array in main.c
  linux-user: Add support for syncfs() syscall
  linux-user: Add support for clock_adjtime() syscall
  linux-user: Fix definition of target_sigevent for 32-bit guests
  linux-user: use libc wrapper instead of direct mremap syscall
  linux-user: Don't use alloca() for epoll_wait's epoll event array
  linux-user: add RTA_PRIORITY in netlink
  linux-user: add kcmp() syscall
  linux-user: sparc64: Use correct target SHMLBA in shmat()
  linux-user: Remove a duplicate item from strace.list
  linux-user: Fix syslog() syscall support
  linux-user: Fix socketcall() syscall support
  linux-user: Fix msgrcv() and msgsnd() syscalls support
  linux-user: Fix mq_open() syscall support
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agolinux-user: disable unicore32 linux-user build
Riku Voipio [Wed, 28 Sep 2016 00:33:23 +0000 (03:33 +0300)]
linux-user: disable unicore32 linux-user build

In order to cleanup linux-user, we need support for most relatively
modern syscalls. unicore32 lacks support for syscalls like
epoll_pwait, preventing cleaning up the CONFIG_EPOLL mess.

This patch can be reverted when unicore32 starts either supporting
the syscalls as defined in mainline kernel, or the oldabi interface
gains support for syscalls supported since at kernel 2.6.19 / glibc 2.6

Cc: MPRC <zhangheng@mprc.pku.edu.cn>
Cc: Xuetao Guan <gxt@mprc.pku.edu.cn>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: added support for pwritev() system call.
Dejan Jovicevic [Tue, 11 Oct 2016 09:52:47 +0000 (11:52 +0200)]
linux-user: added support for pwritev() system call.

This system call performs the same task as the writev() system call,
with the exception of having the fourth argument, offset, which
specifes the file offset at which the input operation is to be performed.
Because of this, the pwritev() implementation is based on the writev()
implementation in linux-user mode.

But, since pwritev() is implemented in the kernel as a 5-argument syscall,
5 arguments are needed to be handled as input and passed to the host
syscall.

The pos_l and pos_h argument of the safe_pwritev() are of type unsigned
long, which can be of different sizes on different platforms. The input
arguments are converted to the appropriate host size when passed to
safe_pwritev().

Signed-off-by: Dejan Jovicevic <dejan.jovicevic@rt-rk.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: added support for preadv() system call.
Dejan Jovicevic [Tue, 11 Oct 2016 09:52:46 +0000 (11:52 +0200)]
linux-user: added support for preadv() system call.

This system call performs the same task as the readv() system call,
with the exception of having the fourth argument, offset, which
specifes the file offset at which the input operation is to be performed.
Because of this, the preadv() implementation is based on the readv()
implementation in linux-user mode.

But, since preadv() is implemented in the kernel as a 5-argument syscall,
5 arguments are needed to be handled as input and passed to the host
syscall.

The pos_l and pos_h argument of the safe_preadv() are of type unsigned
long, which can be of different sizes on different platforms. The input
arguments are converted to the appropriate host size when passed to
safe_preadv().

Signed-off-by: Dejan Jovicevic <dejan.jovicevic@rt-rk.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix fadvise64() syscall support for Mips32
Aleksandar Markovic [Wed, 12 Oct 2016 12:30:25 +0000 (14:30 +0200)]
linux-user: Fix fadvise64() syscall support for Mips32

By looking at the file arch/mips/kernel/scall32-o32.S in Linux
kernel, it can be deduced that, for Mips32 platform, syscall
corresponding to number _NR_fadvise64 as defined in kernel file
arch/mips/include/uapi/asm/unistd.h translates to kernel function
sys_fadvise64_64, and that argument layout for this system call is
as follows:

              0             32 0             32
             +----------------+----------------+
      (arg1) |       fd       |     __pad      | (arg2)
             +----------------+----------------+
      (arg3) |             buffer              | (arg4)
             +----------------+----------------+
      (arg5) |               len               | (arg6)
             +----------------+----------------+
      (arg7) |     advise     |    not used    | (arg8)
             +----------------+----------------+

The same argument layout can be deduced from glibc code, and
relevant commit messages in linux kernel and glibc.

The fix is to change TARGET_NR_fadvise64 to TARGET_NR_fadvise64_64
in Mips32 syscall numbers table. Array mips_syscall_args[] in
linux-user/main.c also already have "fadvise64_64" (and not
"fadvise64") in corresponding place for the syscall number in
question, so no change for linux-user/main.c.

This patch also fixes the failure LTP test posix_fadvise03, if
executed on Qemu-emulated Mips32 platform (user mode).

Signed-off-by: Aleksandar Rikalo <aleksandar.rikalo@imgtec.com>
Signed-off-by: Miroslav Tisma <miroslav.tisma@imgtec.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Redirect termbits.h for Mips64 to termbits.h for Mips32
Aleksandar Markovic [Wed, 12 Oct 2016 12:30:24 +0000 (14:30 +0200)]
linux-user: Redirect termbits.h for Mips64 to termbits.h for Mips32

linux-user/mips64/termbits.h and linux-user/mips/termbits.h
originate from the same files in Linux kernel. There is no plan
to split original headers in Linux kernel into Mips32 and Mips64
versions any time soon. Therefore, it is better not to have
separate Mips32 and Mips64 variants in Qemu.

This patch makes these two files effectively the same, allowing the
mainenance by changing only a single file. (This is already done in
the same fashion for some other headers in same directories.)

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Update ioctls definitions for Mips32
Aleksandar Markovic [Wed, 12 Oct 2016 12:30:23 +0000 (14:30 +0200)]
linux-user: Update ioctls definitions for Mips32

Update linux-user/mips/termbits.h with ioctl definitions from kernel
file arch/mips/include/uapi/asm/ioctls.h.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Update mips_syscall_args[] array in main.c
Aleksandar Markovic [Wed, 12 Oct 2016 12:30:22 +0000 (14:30 +0200)]
linux-user: Update mips_syscall_args[] array in main.c

Array mips_syscall_args[] determines number of arguments for each
syscall on Mips32. It wasn't updated with newer syscalls. Also,
preadv and pwritev have 5 arguments, not 6.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Add support for syncfs() syscall
Aleksandar Markovic [Mon, 10 Oct 2016 11:23:30 +0000 (13:23 +0200)]
linux-user: Add support for syncfs() syscall

This patch implements Qemu user mode syncfs() syscall support. Syscall
syncfs() syncs the filesystem containing file determined by the open
file descriptor passed as the argument to syncfs().

The implementation consists of a straightforward invocation of host's
syncfs(). Configure and strace support is included as well.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Add support for clock_adjtime() syscall
Aleksandar Markovic [Mon, 10 Oct 2016 11:23:29 +0000 (13:23 +0200)]
linux-user: Add support for clock_adjtime() syscall

This patch implements Qemu user mode clock_adjtime() syscall support.

The implementation is based on invocation of host's clock_adjtime().

Signed-off-by: Aleksandar Rikalo <aleksandar.rikalo@imgtec.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix definition of target_sigevent for 32-bit guests
Peter Maydell [Fri, 2 Sep 2016 17:40:01 +0000 (18:40 +0100)]
linux-user: Fix definition of target_sigevent for 32-bit guests

The sigevent structure includes a union with some fields which
are pointers. For the QEMU target_sigevent structure we must
represent these as abi_ulongs, not host function pointers.

This error was causing the compiler to believe it should 8-align
the _sigev_un union on a 64-bit host, which meant that the
code in target_to_host_sigevent() was looking at the wrong
offset to find the _tid field, and timer_create() would
spuriously fail with EINVAL.

This fixes the final loose end noted in LP:1042388.

While we're editing the structure, switch the 'int32_t' fields
to 'abi_int'; this will only matter for guests with non-standard
integer alignment like m68k.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: use libc wrapper instead of direct mremap syscall
Felix Janda [Fri, 30 Sep 2016 23:39:27 +0000 (19:39 -0400)]
linux-user: use libc wrapper instead of direct mremap syscall

This commit essentially reverts commit
3af72a4d98dca033492102603734cbc63cd2694a, which has replaced
five-argument calls to mremap() by direct mremap syscalls for
compatibility with glibc older than version 2.4.

The direct syscall was buggy for 64bit targets on 32bit hosts
because of the default integer type promotions. Since glibc-2.4
is now a decade old, we can remove this workaround.

Signed-off-by: Felix Janda <felix.janda@posteo.de>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Don't use alloca() for epoll_wait's epoll event array
Peter Maydell [Mon, 18 Jul 2016 14:36:00 +0000 (15:36 +0100)]
linux-user: Don't use alloca() for epoll_wait's epoll event array

The epoll event array which epoll_wait() allocates has a size
determined by the guest which could potentially be quite large.
Use g_try_new() rather than alloca() so that we can fail more
cleanly if the guest hands us an oversize value. (ENOMEM is
not a documented return value for epoll_wait() but in practice
some kernel configurations can return it -- see for instance
sys_oabi_epoll_wait() on ARM.)

This rearrangement includes fixing a bug where we were
incorrectly passing a negative length to unlock_user() in
the error-exit codepath.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: add RTA_PRIORITY in netlink
Laurent Vivier [Thu, 6 Oct 2016 12:55:10 +0000 (14:55 +0200)]
linux-user: add RTA_PRIORITY in netlink

Used by fedora21 on ppc64 in the network initialization

Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: add kcmp() syscall
Laurent Vivier [Sun, 25 Sep 2016 20:20:20 +0000 (22:20 +0200)]
linux-user: add kcmp() syscall

Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: sparc64: Use correct target SHMLBA in shmat()
Peter Maydell [Tue, 4 Oct 2016 13:13:46 +0000 (14:13 +0100)]
linux-user: sparc64: Use correct target SHMLBA in shmat()

In commit 40df8c0c0722 support was added for target-specific
handling of SHMLBA. Unfortunately the sparc64-specific part
of the change got lost somewhere between the patch being
posted to the list and going into master:
 http://patchwork.ozlabs.org/patch/646980/
 http://patchwork.ozlabs.org/patch/673339/

Add the accidentally-dropped code.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Remove a duplicate item from strace.list
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:59 +0000 (18:56 +0200)]
linux-user: Remove a duplicate item from strace.list

There is a duplicate item in strace.list. It is benign, but it
shouldn't be there, since it may lead to confusion and even bugs
in the future. It is the only duplicate in strace.list. This
patch removes it.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix syslog() syscall support
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:58 +0000 (18:56 +0200)]
linux-user: Fix syslog() syscall support

There are currently several problems related to syslog() support.

For example, if the second argument "bufp" of target syslog() syscall
is NULL, the current implementation always returns error code EFAULT.
However, NULL is a perfectly valid value for the second argument for
many use cases of this syscall. This is, for example, visible from
this excerpt of man page for syslog(2):

> EINVAL Bad arguments (e.g., bad type; or for type 2, 3, or 4, buf is
>        NULL, or len is less than zero; or for type 8, the level is
>        outside the range 1 to 8).

Moreover, the argument "bufp" is ignored for all cases of values of the
first argument, except 2, 3 and 4. This means that for such cases
(the first argument is not 2, 3 or 4), there is no need to pass "buf"
between host and target, and it can be set to NULL while calling host's
syslog(), without loss of emulation accuracy.

Note also that if "bufp" is NULL and the first argument is 2, 3 or 4, the
correct returned error code is EINVAL, not EFAULT.

All these details are reflected in this patch.

"#ifdef TARGET_NR_syslog" is also proprerly inserted when needed.

Support for Qemu's "-strace" switch for syslog() syscall is included too.

LTP tests syslog11 and syslog12 pass with this patch (while fail without
it), on any platform.

Changes to original patch by Riku Voipio:

 fixed error paths in TARGET_SYSLOG_ACTION_READ_ALL to match

http://lxr.free-electrons.com/source/kernel/printk/printk.c?v=4.7#L1335

Should fix also the build error in:

https://lists.gnu.org/archive/html/qemu-devel/2016-10/msg03721.html

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix socketcall() syscall support
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:57 +0000 (18:56 +0200)]
linux-user: Fix socketcall() syscall support

Since not all Linux host platforms support socketcall() (most notably
Intel), do_socketcall() function in Qemu's syscalls.c is implemented to
mirror the corespondant implementation of socketcall() in Linux kernel,
and to utilise individual socket operations that are supported on all
Linux platforms. (see kernel source file net/socket.c, definition of
socketcall).

However, error codes produced by Qemu implementation are wrong for the
cases of invalid values of the first argument. Also, naming of constants
is not consistent with kernel one, and not consistant with Qemu convention
of prefixing such constants with "TARGET_". This patch in that light
brings do_socketcall() closer to its kernel counterpart, and in that way
fixes the errors and yields more consisrtent Qemu code.

There were also three missing cases (among 20) for strace support for
socketcall(). The array that contains pointers for appropriate printing
functions is updated with 3 elements, however pointers to functions are
left NULL, and its implementation is left for future.

Also, this patch fixes failure of LTP test socketcall02, if executed on some
Qemu emulated sywstems (uer mode).

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix msgrcv() and msgsnd() syscalls support
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:56 +0000 (18:56 +0200)]
linux-user: Fix msgrcv() and msgsnd() syscalls support

If syscalls msgrcv() and msgsnd() fail, they return E2BIG, EACCES,
EAGAIN, EFAULT, EIDRM, EINTR, EINVAL, ENOMEM, or ENOMSG.

By examining negative scenarios of these syscalls for Mips, it was
established that ENOMSG does not have the same value accross all
platforms, but it is nevertheless not included for conversion in
the correspondant conversion table defined in linux-user/syscall.c.
This is certainly a bug, since it leads to the incorrect emulation
of msgrcv() and msgsnd() for scenarios involving ENOMSG.

This patch fixes this by extending the conversion table to include
ENOMSG.

Also, LTP test msgrcv04 will be fixed for some platforms.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Fix mq_open() syscall support
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:55 +0000 (18:56 +0200)]
linux-user: Fix mq_open() syscall support

Conversion of file creation flags (O_CREAT, ...) from target to host
was missing.

Also, this patch implements better error handling.

Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agolinux-user: Add support for adjtimex() syscall
Aleksandar Markovic [Thu, 22 Sep 2016 16:56:50 +0000 (18:56 +0200)]
linux-user: Add support for adjtimex() syscall

This patch implements Qemu user mode adjtimex() syscall support.

Syscall adjtimex() reads and optionally sets parameters for a clock
adjustment algorithm used in network synchonization or similar scenarios.

Its declaration is:

int adjtimex(struct timex *buf);

The correspondent source code in the Linux kernel is at kernel/time.c,
line 206.

The Qemu implementation is based on invocation of host's adjtimex(), and
its key part is in the "TARGET_NR_adjtimex" case segment of the the main
switch statement of the function do_syscall(), in linux-user/syscalls.c. All
necessary conversions of the data structures from target to host and from
host to target are covered. Two new functions, target_to_host_timex() and
host_to_target_timex(), are provided for the purpose of such conversions.
For that purpose, the support for related structure "timex" had tp be added
to the file linux-user/syscall_defs.h, based on its definition in Linux
kernel. Also, the relevant support for "-strace" Qemu option is included
in files linux-user/strace.c and linux-user/strace.list.

This patch also fixes failures of LTP tests adjtimex01 and adjtimex02, if
executed in Qemu user mode.

Signed-off-by: Aleksandar Rikalo <aleksandar.rikalo@imgtec.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/berrange/tags/pull-qcrypto-2016-10-20-1' into...
Peter Maydell [Thu, 20 Oct 2016 13:46:19 +0000 (14:46 +0100)]
Merge remote-tracking branch 'remotes/berrange/tags/pull-qcrypto-2016-10-20-1' into staging

Merge qcrypto 2016/10/20 v1

# gpg: Signature made Thu 20 Oct 2016 12:58:41 BST
# gpg:                using RSA key 0xBE86EBB415104FDF
# gpg: Good signature from "Daniel P. Berrange <dan@berrange.com>"
# gpg:                 aka "Daniel P. Berrange <berrange@redhat.com>"
# Primary key fingerprint: DAF3 A6FD B26B 6291 2D0E  8E3F BE86 EBB4 1510 4FDF

* remotes/berrange/tags/pull-qcrypto-2016-10-20-1:
  crypto: fix initialization of gcrypt threading
  crypto: fix initialization of crypto in tests
  qtest: fix make check complaint in crypto module
  crypto: add mode check in qcrypto_cipher_new() for cipher-builtin
  crypto: add CTR mode support
  crypto: extend mode as a parameter in qcrypto_cipher_supports()

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agocrypto: fix initialization of gcrypt threading
Daniel P. Berrange [Mon, 10 Oct 2016 11:17:50 +0000 (12:17 +0100)]
crypto: fix initialization of gcrypt threading

The gcrypt threads implementation must be set before calling
any other gcrypt APIs, especially gcry_check_version(),
since that triggers initialization of the random pool. After
that is initialized, changes to the threads impl won't be
honoured by the random pool code. This means that gcrypt
will think thread locking is needed and so try to acquire
the random pool mutex, but this is NULL as no threads impl
was set originally. This results in a crash in the random
pool code.

For the same reasons, we must set the gcrypt threads impl
before calling gnutls_init, since that will also trigger
gcry_check_version

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agocrypto: fix initialization of crypto in tests
Daniel P. Berrange [Thu, 29 Sep 2016 11:43:35 +0000 (12:43 +0100)]
crypto: fix initialization of crypto in tests

The test-io-channel-tls test was missing a call to qcrypto_init
and test-crypto-hash was initializing it multiple times,

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agoqtest: fix make check complaint in crypto module
Gonglei [Sat, 24 Sep 2016 00:36:19 +0000 (08:36 +0800)]
qtest: fix make check complaint in crypto module

  CC    tests/test-crypto-tlscredsx509.o
  CC    tests/crypto-tls-x509-helpers.o
  CC    tests/pkix_asn1_tab.o
tests/pkix_asn1_tab.c:7:22: warning: libtasn1.h: No such file or directory
tests/pkix_asn1_tab.c:9: error: expected ‘=’, ‘,’, ‘;’, ‘asm’ or ‘__attribute__’ before ‘pkix_asn1_tab’
make: *** [tests/pkix_asn1_tab.o] Error 1

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agocrypto: add mode check in qcrypto_cipher_new() for cipher-builtin
Gonglei [Mon, 26 Sep 2016 09:23:23 +0000 (17:23 +0800)]
crypto: add mode check in qcrypto_cipher_new() for cipher-builtin

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agocrypto: add CTR mode support
Gonglei [Mon, 26 Sep 2016 09:23:22 +0000 (17:23 +0800)]
crypto: add CTR mode support

Introduce CTR mode support for the cipher APIs.
CTR mode uses a counter rather than a traditional IV.
The counter has additional properties, including a nonce
and initial counter block. We reuse the ctx->iv as
the counter for conveniences.

Both libgcrypt and nettle are support CTR mode, the
cipher-builtin doesn't support yet.

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agocrypto: extend mode as a parameter in qcrypto_cipher_supports()
Gonglei [Mon, 26 Sep 2016 09:23:21 +0000 (17:23 +0800)]
crypto: extend mode as a parameter in qcrypto_cipher_supports()

It can't guarantee all cipher modes are supported
if one cipher algorithm is supported by a backend.
Let's extend qcrypto_cipher_supports() to take both
the algorithm and mode as parameters.

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/awilliam/tags/vfio-updates-20161017.0' into...
Peter Maydell [Tue, 18 Oct 2016 10:40:27 +0000 (11:40 +0100)]
Merge remote-tracking branch 'remotes/awilliam/tags/vfio-updates-20161017.0' into staging

VFIO updates 2016-10-17

 - Convert to realize & improve error reporting (Eric Auger)
 - RTL quirk bug fix (Thorsten Kohfeldt)
 - Skip duplicate pre/post reset (Cao jin)

# gpg: Signature made Mon 17 Oct 2016 20:42:44 BST
# gpg:                using RSA key 0x239B9B6E3BB08B22
# gpg: Good signature from "Alex Williamson <alex.williamson@redhat.com>"
# gpg:                 aka "Alex Williamson <alex@shazbot.org>"
# gpg:                 aka "Alex Williamson <alwillia@redhat.com>"
# gpg:                 aka "Alex Williamson <alex.l.williamson@gmail.com>"
# Primary key fingerprint: 42F6 C04E 540B D1A9 9E7B  8A90 239B 9B6E 3BB0 8B22

* remotes/awilliam/tags/vfio-updates-20161017.0:
  vfio: fix duplicate function call
  vfio/pci: Fix vfio_rtl8168_quirk_data_read address offset
  vfio/pci: Handle host oversight
  vfio/pci: Remove vfio_populate_device returned value
  vfio/pci: Remove vfio_msix_early_setup returned value
  vfio/pci: Conversion to realize
  vfio/platform: Pass an error object to vfio_base_device_init
  vfio/platform: fix a wrong returned value in vfio_populate_device
  vfio/platform: Pass an error object to vfio_populate_device
  vfio: Pass an error object to vfio_get_device
  vfio: Pass an error object to vfio_get_group
  vfio: Pass an Error object to vfio_connect_container
  vfio/pci: Pass an error object to vfio_pci_igd_opregion_init
  vfio/pci: Pass an error object to vfio_add_capabilities
  vfio/pci: Pass an error object to vfio_intx_enable
  vfio/pci: Pass an error object to vfio_msix_early_setup
  vfio/pci: Pass an error object to vfio_populate_device
  vfio/pci: Pass an error object to vfio_populate_vga
  vfio/pci: Use local error object in vfio_initfn

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/ehabkost/tags/machine-pull-request' into staging
Peter Maydell [Tue, 18 Oct 2016 09:33:30 +0000 (10:33 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/machine-pull-request' into staging

machine + memory backend queue, 2016-10-17

# gpg: Signature made Mon 17 Oct 2016 18:54:57 BST
# gpg:                using RSA key 0x2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/machine-pull-request:
  hostmem-file: Register TYPE_MEMORY_BACKEND_FILE properties as class properties
  hostmem: Register TYPE_MEMORY_BACKEND properties as class properties
  pc: Register TYPE_PC_MACHINE properties as class properties
  machine: Register TYPE_MACHINE properties as class properties
  machine: Fix replacement of '_' by '-' in machine property names

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging
Peter Maydell [Tue, 18 Oct 2016 08:29:43 +0000 (09:29 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging

x86 queue, 2016-10-17

# gpg: Signature made Mon 17 Oct 2016 18:51:07 BST
# gpg:                using RSA key 0x2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/x86-pull-request: (21 commits)
  target-i386: Don't use cpu->migratable when filtering features
  target-i386: Return runnability information on query-cpu-definitions
  target-i386: x86_cpu_load_features() function
  target-i386: Unset cannot_destroy_with_object_finalize_yet
  target-i386/kvm: cache the return value of kvm_enable_x2apic()
  intel_iommu: reject broken EIM
  intel_iommu: add OnOffAuto intr_eim as "eim" property
  intel_iommu: redo configuraton check in realize
  intel_iommu: pass whole remapped addresses to apic
  apic: add send_msi() to APICCommonClass
  apic: add global apic_get_class()
  target-i386: Move warning code outside x86_cpu_filter_features()
  qmp: Add runnability information to query-cpu-definitions
  target-i386: xsave: Add FP and SSE bits to x86_ext_save_areas
  target-i386: Register properties for feature aliases manually
  target-i386: Remove underscores from feat_names arrays
  target-i386: Make plus_features/minus_features QOM-based
  target-i386: Register aliases for feature names with underscores
  target-i386: Disable VME by default with TCG
  target-i386: List CPU models using subclass list
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20161017' into...
Peter Maydell [Mon, 17 Oct 2016 18:41:23 +0000 (19:41 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20161017' into staging

target-arm:
 * target-arm: kvm: use AddressSpace-specific listener
 * aspeed: add SMC controllers
 * hw/arm/boot: allow using a command line specified dtb without a kernel
 * hw/dma/pl080: Fix bad bit mask
 * hw/intc/arm_gic_kvm: Fix build on aarch64 with some compilers
 * hw/arm/virt: fix ACPI tables for ITS
 * tests: add a m25p80 test
 * tests: cleanup ptimer-test
 * pxa2xx: Auto-assign name for i2c bus in i2c_init_bus
 * target-arm: handle tagged addresses in A64 code
 * target-arm: Fix masking of PC lower bits when doing exception returns
 * target-arm: Implement dummy MDCCINT_EL1
 * target-arm: Add trace events for the generic timers
 * hw/intc/arm_gicv3: Fix ICC register tracepoints
 * hw/char/pl011: Add trace events

# gpg: Signature made Mon 17 Oct 2016 19:39:42 BST
# gpg:                using RSA key 0x3C2525ED14360CDE
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>"
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>"
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>"
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20161017: (25 commits)
  hw/char/pl011: Add trace events
  hw/intc/arm_gicv3: Fix ICC register tracepoints
  target-arm: Add trace events for the generic timers
  target-arm: Implement dummy MDCCINT_EL1
  Fix masking of PC lower bits when doing exception returns
  target-arm: Comments added to identify cases in a switch
  target-arm: Code changes to implement overwrite of tag field on PC load
  target-arm: Infrastucture changes to enable handling of tagged address loading into PC
  pxa2xx: Auto-assign name for i2c bus in i2c_init_bus.
  tests: cleanup ptimer-test
  tests: add a m25p80 test
  hw/arm/virt: no ITS on older machine types
  hw/arm/virt-acpi-build: fix MADT generation
  hw/intc/arm_gic_kvm: Fix build on aarch64
  hw/dma/pl080: Fix bad bit mask (PL080_CONF_M1 | PL080_CONF_M1)
  hw/arm/boot: allow using a command line specified dtb without a kernel
  aspeed: add support for the SMC segment registers
  aspeed: create mapping regions for the maximum number of slaves
  aspeed: add support for the AST2500 SoC SMC controllers
  aspeed: extend the number of host SPI controllers
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/char/pl011: Add trace events
Peter Maydell [Wed, 12 Oct 2016 17:54:36 +0000 (18:54 +0100)]
hw/char/pl011: Add trace events

Add some trace events for the pl011 UART model.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1476294876-12340-5-git-send-email-peter.maydell@linaro.org

8 years agohw/intc/arm_gicv3: Fix ICC register tracepoints
Peter Maydell [Wed, 12 Oct 2016 17:54:35 +0000 (18:54 +0100)]
hw/intc/arm_gicv3: Fix ICC register tracepoints

Fix some problems with the tracepoints for ICC register reads
and writes:
 * tracepoints for ICC_BPR<n>, ICC_AP<n>R<x>, ICC_IGRPEN<n>,
   ICC_EIOR<n> were not printing the <n> that indicated whether
   the access was to the group 0 or 1 register
 * the ICC_IGREPEN1_EL3 read function was not actually calling
   the associated tracepoint
 * the ICC_BPR<n> write function was incorrectly calling the
   tracepoint for ICC_PMR writes

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Message-id: 1476294876-12340-4-git-send-email-peter.maydell@linaro.org

8 years agotarget-arm: Add trace events for the generic timers
Peter Maydell [Wed, 12 Oct 2016 17:54:34 +0000 (18:54 +0100)]
target-arm: Add trace events for the generic timers

Add some useful trace events for the ARM generic timers (notably
the various register writes and the resulting IRQ line state).

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Message-id: 1476294876-12340-3-git-send-email-peter.maydell@linaro.org

8 years agotarget-arm: Implement dummy MDCCINT_EL1
Peter Maydell [Wed, 12 Oct 2016 17:54:33 +0000 (18:54 +0100)]
target-arm: Implement dummy MDCCINT_EL1

MDCCINT_EL1 is part of the DCC debugger communication
channel between the CPU and an attached external debugger.
QEMU doesn't implement this, but since Linux may try
to access this register we need to provide at least
a dummy implementation.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Message-id: 1476294876-12340-2-git-send-email-peter.maydell@linaro.org

8 years agoFix masking of PC lower bits when doing exception returns
Peter Maydell [Mon, 10 Oct 2016 15:26:03 +0000 (16:26 +0100)]
Fix masking of PC lower bits when doing exception returns

In commit 9b6a3ea7a699594 store_reg() was changed to mask
both bits 0 and 1 of the new PC value when in ARM mode.
Unfortunately this broke the exception return code paths
when doing a return from ARM mode to Thumb mode: in some
of these we write a new CPSR including new Thumb mode
bit via gen_helper_cpsr_write_eret(), and then use store_reg()
to write the new PC. In this case if the new CPSR specified
Thumb mode then masking bit 1 of the PC is incorrect
(these code paths correspond to the v8 ARM ARM pseudocode
function AArch32.ExceptionReturn(), which always aligns the
new PC appropriately for the new instruction set state).

Instead of using store_reg() in exception-return code paths,
call a new store_pc_exc_ret() which stores the raw new PC
value to env->regs[15], and then mask it appropriately in
the subsequent helper_cpsr_write_eret() where the new
env->thumb state is available.

This fixes a bug introduced by 9b6a3ea7a699594 which caused
crashes/hangs or otherwise bad behaviour for Linux when
userspace was using Thumb.

Reported-by: Jerome Forissier <jerome.forissier@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1476113163-24578-1-git-send-email-peter.maydell@linaro.org

8 years agotarget-arm: Comments added to identify cases in a switch
Thomas Hanson [Mon, 17 Oct 2016 18:22:18 +0000 (19:22 +0100)]
target-arm: Comments added to identify cases in a switch

3 cases in a switch in disas_exc() require reference to the
ARM ARM spec in order to determine what case they're handling.

Signed-off-by: Thomas Hanson <thomas.hanson@linaro.org>
Message-id: 1476301853-15774-5-git-send-email-thomas.hanson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotarget-arm: Code changes to implement overwrite of tag field on PC load
Thomas Hanson [Mon, 17 Oct 2016 18:22:18 +0000 (19:22 +0100)]
target-arm: Code changes to implement overwrite of tag field on PC load

For BR, BLR and RET instructions, if tagged addresses are enabled, the
tag field in the address must be cleared out prior to loading the
address into the PC.  Depending on the current EL, it will be set to
either all 0's or all 1's.

Signed-off-by: Thomas Hanson <thomas.hanson@linaro.org>
Message-id: 1476301853-15774-3-git-send-email-thomas.hanson@linaro.org
[PMM: remove unnecessary gen_a64_set_pc_reg() wrapper,
 rename gen_a64_set_pc_var() to gen_a64_set_pc(), fix stray
 misindentation]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotarget-arm: Infrastucture changes to enable handling of tagged address loading into PC
Thomas Hanson [Mon, 17 Oct 2016 18:22:18 +0000 (19:22 +0100)]
target-arm: Infrastucture changes to enable handling of tagged address loading into PC

When capturing the current CPU state for the TB, extract the TBI0 and TBI1
values from the correct TCR for the current EL and then add them to the TB
flags field.

Then, at the start of code generation for the block, copy the TBI fields
into the DisasContext structure.

Signed-off-by: Thomas Hanson <thomas.hanson@linaro.org>
Message-id: 1476301853-15774-2-git-send-email-thomas.hanson@linaro.org
[PMM: drop useless 'extern' keyword on function prototypes;
 provide CONFIG_USER_ONLY trivial versions of arm_regime_tbi[01]()]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agopxa2xx: Auto-assign name for i2c bus in i2c_init_bus.
Vijay Kumar B [Mon, 17 Oct 2016 18:22:18 +0000 (19:22 +0100)]
pxa2xx: Auto-assign name for i2c bus in i2c_init_bus.

If a name is provided, the same name is assigned to both the I2C
controllers. Leaving it NULL, causes names to be automatically
assigned with an ID suffix, giving unique names to each
controller. This helps us to uniquely identify each controller in the
device tree, for example when adding an I2C device.

Signed-off-by: Vijay Kumar B. <vijaykumar@zilogic.com>
Reviewed-by: Deepak S. <deepak@zilogic.com>
Message-id: 1476351885-8905-1-git-send-email-vijaykumar@zilogic.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotests: cleanup ptimer-test
Paolo Bonzini [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
tests: cleanup ptimer-test

1) ptimer-test is not a qtest---it runs the ptimer.c code directly in the
ptimer-test process

2) ptimer-test has its own stubs file, so there is no need to add more
stubs to stubs/vmstate.c

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotests: add a m25p80 test
Cédric Le Goater [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
tests: add a m25p80 test

This test uses the palmetto platform and the Aspeed SPI controller to
test the m25p80 flash module device model. The flash model is defined
by the platform (n25q256a) and it would be nice to find way to control
it, using a property probably.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1475787271-28794-1-git-send-email-clg@kaod.org
Brainstormed-with: Greg Kurz <groug@kaod.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/arm/virt: no ITS on older machine types
Andrew Jones [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
hw/arm/virt: no ITS on older machine types

We should avoid exposing new hardware (through DT and ACPI) on older
machine types. This patch keeps 2.7 and older from changing, despite
the introduction of ITS support for 2.8.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Message-id: 1476117341-32690-3-git-send-email-drjones@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/arm/virt-acpi-build: fix MADT generation
Andrew Jones [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
hw/arm/virt-acpi-build: fix MADT generation

We can't return early from build_* functions, as build_header is
only called at the end.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Message-id: 1476117341-32690-2-git-send-email-drjones@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/intc/arm_gic_kvm: Fix build on aarch64
Christopher Covington [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
hw/intc/arm_gic_kvm: Fix build on aarch64

Remove unused debugging code to fix native building on aarch64. Without
this change, the following -Werr output inhibits make from completing.

  qemu/hw/intc/arm_gic_kvm.c:38:18: error: debug_gic_kvm defined but not used [-Werror=unused-const-variable=]
   static const int debug_gic_kvm = 0;
                    ^~~~~~~~~~~~~
  cc1: all warnings being treated as errors
  qemu/rules.mak:60: recipe for target 'hw/intc/arm_gic_kvm.o' failed
  make[1]: *** [hw/intc/arm_gic_kvm.o] Error 1
  Makefile:205: recipe for target 'subdir-aarch64-softmmu' failed

Signed-off-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20161011163202.19720-1-cov@codeaurora.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/dma/pl080: Fix bad bit mask (PL080_CONF_M1 | PL080_CONF_M1)
Thomas Huth [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
hw/dma/pl080: Fix bad bit mask (PL080_CONF_M1 | PL080_CONF_M1)

The M1 and M2 bits are both used for configuring the endianness
of the AHB master interfaces, so the second PL080_CONF_M1 should
be PL080_CONF_M2 instead.

Buglink: https://bugs.launchpad.net/qemu/+bug/1631773
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-id: 1476274451-26567-1-git-send-email-thuth@redhat.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/arm/boot: allow using a command line specified dtb without a kernel
Michael Olbrich [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
hw/arm/boot: allow using a command line specified dtb without a kernel

When kernel and device tree are specified in the QEMU commandline, then
this device tree may be modified e.g. to add virtio_mmio devices.
With a bootloader e.g. on a flash device these extra devices are not
available.
With this change, the device tree can be specified at the QEMU commandline.
The modified device tree made available to the bootloader with the same
mechanism already supported by device trees fully generated by QEMU.

Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
Message-id: 1473520054-402-1-git-send-email-m.olbrich@pengutronix.de
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: add support for the SMC segment registers
Cédric Le Goater [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
aspeed: add support for the SMC segment registers

The SMC controller on the Aspeed SoC has a set of registers to
configure the mapping of each flash module in the SoC address
space. Writing to these registers triggers a remap of the memory
region and the spec requires a certain number of checks before doing
so.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-7-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: create mapping regions for the maximum number of slaves
Cédric Le Goater [Mon, 17 Oct 2016 18:22:17 +0000 (19:22 +0100)]
aspeed: create mapping regions for the maximum number of slaves

The SMC controller on the Aspeed SoC has a set of registers to
configure the mapping of each flash module in the SoC address
space. These mapping windows are configurable even though no SPI slave
is attached to the controller.

Also rewrite a bit the comments in the code on this topic.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-6-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: add support for the AST2500 SoC SMC controllers
Cédric Le Goater [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
aspeed: add support for the AST2500 SoC SMC controllers

The SMC controllers on the Aspeed AST2500 SoC are very similar to the
ones found on the AST2400. The differences are on the number of
supported flash modules and their default mappings in the SoC address
space.

The Aspeed AST2500 has one SPI controller for the BMC firmware and two
for the host firmware. All controllers have now the same set of
registers compatible with the AST2400 FMC controller and the legacy
'SMC' controller is fully gone.

We keep the FMC object to act as the BMC SPI controller and add a new
SPI controller for the host. We also have to introduce new type names
to handle the differences in the flash modules memory mappping.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-5-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: extend the number of host SPI controllers
Cédric Le Goater [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
aspeed: extend the number of host SPI controllers

The AST2500 SoC has two. Let's prepare ground for the next changes
which will add the required definitions for the second host SPI
controller.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-4-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: move the flash module mapping address under the controller definition
Cédric Le Goater [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
aspeed: move the flash module mapping address under the controller definition

This will ease the definition of the new controllers for the AST2500
SoC and also ease the support of the segment registers, which provide
a way to reconfigure the mapping window of each slave.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-3-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoaspeed: rename the smc object to fmc
Cédric Le Goater [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
aspeed: rename the smc object to fmc

The Aspeed SoC has three different types of SMC (Static Memory
Controller) controllers: the SMC (legacy), the FMC (the new one) and
the SPI for the host PNOR. The FMC and the SPI models are now
converging on the AST2500 SoC and the SMC, which was still available
on the AST2400 SoC, was removed.

The Aspeed SoC does not provide support for the legacy SMC
controller. So, let's rename the 'smc' object to 'fmc' to clarify its
nature.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1474977462-28032-2-git-send-email-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotarget-arm: kvm: use AddressSpace-specific listener
Paolo Bonzini [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
target-arm: kvm: use AddressSpace-specific listener

The only address space where the GIC devices are added is
address_space_memory.  There is no need to use a global
MemoryListener.

This removes the only user of global MemoryListeners.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
[PMM: added missing #include "exec/address-spaces.h"]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1475219846-32609-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoReducing stack frame size in stream_process_mem2s()
Rutuja Shah [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
Reducing stack frame size in stream_process_mem2s()

This patch allocates memory for txbuf in struct Stream rather than the stack.
As a result, the stack frame size is reduced of stream_process_mem2s().

Signed-off-by: Rutuja Shah <rutu.shah.26@gmail.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@xilinx.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agodocs/generic-loader: Update the document
Alistair Francis [Mon, 17 Oct 2016 18:22:16 +0000 (19:22 +0100)]
docs/generic-loader: Update the document

This patch does three things:
 - It adds a list of restrictions and ToDos
 - It corrects the header --- lines to match the length of the header
 - It clarifies the force-raw option

Signed-off-by: Alistair Francis <alistair.francis@xilinx.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-id: e75d1d285cf8f45037c41ebe1bc3f68120f09cb9.1475702918.git.alistair.francis@xilinx.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotarget-i386: Don't use cpu->migratable when filtering features
Eduardo Habkost [Fri, 14 Oct 2016 18:42:45 +0000 (15:42 -0300)]
target-i386: Don't use cpu->migratable when filtering features

When explicitly enabling unmigratable flags using "-cpu host"
(e.g. "-cpu host,+invtsc"), the requested feature won't be
enabled because cpu->migratable is true by default.

This is inconsistent with all other CPU models, which don't have
the "migratable" option, making "+invtsc" work without the need
for extra options.

This happens because x86_cpu_filter_features() uses
cpu->migratable as an argument for
x86_cpu_get_supported_feature_word(). This is not useful
because:
2) on "-cpu host" it only makes QEMU disable features that were
   explicitly enabled in the command-line;
1) on all the other CPU models, cpu->migratable is already false.

The fix is to just use 'false' as an argument to
x86_cpu_get_supported_feature_word() in
x86_cpu_filter_features().

Note that:

* This won't change anything for people using using
  "-cpu host" or "-cpu host,migratable=<on|off>" (with no extra
  features) because the x86_cpu_get_supported_feature_word() call
  on the cpu->host_features check uses cpu->migratable as
  argument.
* This won't change anything for any CPU model except "host"
  because they all have cpu->migratable == false (and only "host"
  has the "migratable" property that allows it to be changed).
* This will only change things for people using "-cpu host,+<feature>",
  where <feature> is a non-migratable feature. The only existing
  named non-migratable feature is "invtsc".

In other words, this change will only affect people using
"-cpu host,+invtsc" (that will now get what they asked for: the
invtsc flag will be enabled). All other use cases are unaffected.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agohostmem-file: Register TYPE_MEMORY_BACKEND_FILE properties as class properties
Eduardo Habkost [Thu, 13 Oct 2016 21:18:41 +0000 (18:18 -0300)]
hostmem-file: Register TYPE_MEMORY_BACKEND_FILE properties as class properties

To do the conversion, the file_backend_class_init() was moved
after the getter/setter functions. The old
file_backend_instance_init() function was removed because it is
not needed anymore.

The NULL errp arguments on the property registration calls were
changed to &error_abort.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agohostmem: Register TYPE_MEMORY_BACKEND properties as class properties
Eduardo Habkost [Thu, 13 Oct 2016 20:52:51 +0000 (17:52 -0300)]
hostmem: Register TYPE_MEMORY_BACKEND properties as class properties

The NULL errp arguments on the property registration calls were
changed to &error_abort.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agopc: Register TYPE_PC_MACHINE properties as class properties
Eduardo Habkost [Thu, 13 Oct 2016 20:48:36 +0000 (17:48 -0300)]
pc: Register TYPE_PC_MACHINE properties as class properties

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agomachine: Register TYPE_MACHINE properties as class properties
Eduardo Habkost [Thu, 13 Oct 2016 20:40:46 +0000 (17:40 -0300)]
machine: Register TYPE_MACHINE properties as class properties

When doing the conversion, the NULL errp arguments on the
property registration calls were changed to &error_abort.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agomachine: Fix replacement of '_' by '-' in machine property names
Markus Armbruster [Thu, 13 Oct 2016 16:44:14 +0000 (18:44 +0200)]
machine: Fix replacement of '_' by '-' in machine property names

machine_set_property() replaces '_' by '-' in the property name.
Except it fails to replace an initial '_'.  Screwed up in commit
b0ddb8b.  Reproducer: "-M pc,__foo_bar=true" produces "Property
'._-foo-bar' not found".

Error messages using a mangled name rather than the name the user
actually wrote is user-hostile, but that's a different topic.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386: Return runnability information on query-cpu-definitions
Eduardo Habkost [Tue, 3 May 2016 16:48:13 +0000 (13:48 -0300)]
target-i386: Return runnability information on query-cpu-definitions

Fill the "unavailable-features" field on the x86 implementation
of query-cpu-definitions.

Cc: Jiri Denemark <jdenemar@redhat.com>
Cc: libvir-list@redhat.com
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386: x86_cpu_load_features() function
Eduardo Habkost [Wed, 28 Sep 2016 18:03:26 +0000 (15:03 -0300)]
target-i386: x86_cpu_load_features() function

When probing for CPU model information, we need to reuse the code
that initializes CPUID fields, but not the remaining side-effects
of x86_cpu_realizefn(). Move that code to a separate function
that can be reused later.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386: Unset cannot_destroy_with_object_finalize_yet
Eduardo Habkost [Mon, 10 Oct 2016 19:31:45 +0000 (16:31 -0300)]
target-i386: Unset cannot_destroy_with_object_finalize_yet

TYPE_X86_CPU now call cpu_exec_init() on realize, so we don't
need to set cannot_destroy_with_object_finalize_yet anymore.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386/kvm: cache the return value of kvm_enable_x2apic()
Radim Krčmář [Mon, 10 Oct 2016 15:28:48 +0000 (17:28 +0200)]
target-i386/kvm: cache the return value of kvm_enable_x2apic()

Assume that KVM would have returned the same on subsequent runs.
Abstract the memoizaiton pattern into macros and call it memorize as
adding the r makes it less obscure.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agointel_iommu: reject broken EIM
Radim Krčmář [Mon, 10 Oct 2016 15:28:47 +0000 (17:28 +0200)]
intel_iommu: reject broken EIM

Cluster x2APIC cannot work without KVM's x2apic API when the maximal
APIC ID is greater than 8 and only KVM's LAPIC can support x2APIC, so we
forbid other APICs and also the old KVM case with less than 9, to
simplify the code.

There is no point in enabling EIM in forbidden APICs, so we keep it
enabled only for the KVM APIC;  unconditionally, because making the
option depend on KVM version would be a maintanance burden.

Old QEMUs would enable eim whenever intremap was on, which would trick
guests into thinking that they can enable cluster x2APIC even if any
interrupt destination would get clamped to 8 bits.
Depending on your configuration, QEMU could notice that the destination
LAPIC is not present and report it with a very non-obvious:

  KVM: injection failed, MSI lost (Operation not permitted)

Or the guest could say something about unexpected interrupts, because
clamping leads to aliasing so interrupts were being delivered to
incorrect VCPUs.

KVM_X2APIC_API is the feature that allows us to enable EIM for KVM.

QEMU 2.7 allowed EIM whenever interrupt remapping was enabled.  In order
to keep backward compatibility, we again allow guests to misbehave in
non-obvious ways, and make it the default for old machine types.

A user can enable the buggy mode it with "x-buggy-eim=on".

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agointel_iommu: add OnOffAuto intr_eim as "eim" property
Radim Krčmář [Mon, 10 Oct 2016 15:28:46 +0000 (17:28 +0200)]
intel_iommu: add OnOffAuto intr_eim as "eim" property

The default (auto) emulates the current behavior.
A user can now control EIM like
  -device intel-iommu,intremap=on,eim=off

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agointel_iommu: redo configuraton check in realize
Radim Krčmář [Mon, 10 Oct 2016 15:28:45 +0000 (17:28 +0200)]
intel_iommu: redo configuraton check in realize

* there no point in configuring the device if realization is going to
  fail, so move the check to the beginning,
* create a separate function for the check,
* use error_setg() instead error_report().

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agointel_iommu: pass whole remapped addresses to apic
Radim Krčmář [Mon, 10 Oct 2016 15:28:44 +0000 (17:28 +0200)]
intel_iommu: pass whole remapped addresses to apic

The MMIO interface to APIC only allowed 8 bit addresses, which is not
enough for 32 bit addresses from EIM remapping.
Intel stored upper 24 bits in the high MSI address, so use the same
technique. The technique is also used in KVM MSI interface.
Other APICs are unlikely to handle those upper bits.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agoapic: add send_msi() to APICCommonClass
Radim Krčmář [Mon, 10 Oct 2016 15:28:43 +0000 (17:28 +0200)]
apic: add send_msi() to APICCommonClass

The MMIO based interface to APIC doesn't work well with MSIs that have
upper address bits set (remapped x2APIC MSIs).  A specialized interface
is a quick and dirty way to avoid the shortcoming.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agoapic: add global apic_get_class()
Radim Krčmář [Mon, 10 Oct 2016 15:28:42 +0000 (17:28 +0200)]
apic: add global apic_get_class()

Every configuration has only up to one APIC class and we'll be extending
the class with a function that can be called without an instanced
object, so a direct access to the class is convenient.

This patch will break compilation if some code uses apic_get_class()
with CONFIG_USER_ONLY.

Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386: Move warning code outside x86_cpu_filter_features()
Eduardo Habkost [Wed, 4 May 2016 17:50:46 +0000 (14:50 -0300)]
target-i386: Move warning code outside x86_cpu_filter_features()

x86_cpu_filter_features() will be reused by code that shouldn't
print any warning. Move the warning code to a new
x86_cpu_report_filtered_features() function, and call it from
x86_cpu_realizefn().

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agoqmp: Add runnability information to query-cpu-definitions
Eduardo Habkost [Fri, 30 Sep 2016 18:49:45 +0000 (15:49 -0300)]
qmp: Add runnability information to query-cpu-definitions

Add a new optional field to query-cpu-definitions schema:
"unavailable-features". It will contain a list of QOM properties
that prevent the CPU model from running in the current host.

Cc: David Hildenbrand <dahi@linux.vnet.ibm.com>
Cc: Michael Mueller <mimu@linux.vnet.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Jiri Denemark <jdenemar@redhat.com>
Cc: libvir-list@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agotarget-i386: xsave: Add FP and SSE bits to x86_ext_save_areas
Eduardo Habkost [Fri, 30 Sep 2016 18:49:42 +0000 (15:49 -0300)]
target-i386: xsave: Add FP and SSE bits to x86_ext_save_areas

Instead of treating the FP and SSE bits as special cases, add
them to the x86_ext_save_areas array. This will simplify the code
that calculates the supported xsave components and the size of
the xsave area.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>